Meeting Spotlight

Meeting Spotlight: Don't Re-watch, Just Relive.

Comment

Inspiration Office professionals waste hours scrubbing through long meeting recordings to find specific moments or decisions. We wanted to make video as searchable as text, allowing users to find the "point of action" instantly without watching the entire file.

What it does Meeting Spotlight is a multimodal navigator that lets users find topics in videos using voice or text. Powered by Gemini 3, it analyzes meeting footage to identify precise timestamps and provides a "Spotlight Preview" button to jump directly to that exact second in the player.

How we built it We used Flutter for the cross-platform frontend and the Gemini 3 Pro API for deep video reasoning. By leveraging Gemini’s 1-million-token context window and High Thinking level, we bypassed the need for traditional transcription, processing raw video and audio queries directly.

Challenges we ran into Achieving second-level accuracy was difficult. We overcame this by refining our system instructions to include a temporal buffer and utilized the Gemini File API to manage large meeting files efficiently without hitting payload limits.

Accomplishments that we're proud of We successfully implemented a "zero-transcript" search, where the AI "watches" the video natively. We are also proud of the seamless integration between the voice-query recorder and the video player’s automated seeking logic.

What we learned We learned that Gemini 3 is highly effective at temporal grounding—the ability to map concepts to specific timeframes. We also discovered that providing the model with a clear persona as a "Temporal Analyst" significantly improved the reliability of our JSON outputs.

What's next for Meeting Spotlight The next steps involve cross-meeting search to track project evolution over time and direct integration with corporate calendars to automatically import and index every meeting as it concludes.

Built With

Share this project:

Updates